Singapore Model AI Governance Framework

What it is

The Model AI Governance Framework is Singapore’s practical guide to deploying AI responsibly. The PDPC released the first edition on 23 January 2019 at the World Economic Forum in Davos. The second edition followed on 21 January 2020, also at Davos, refining the first with industry examples, clearer guidance on human involvement and new measures for robustness, reproducibility and auditability.

The framework rests on two guiding principles. Organisations using AI in decision-making should make sure the process is explainable, transparent and fair. And AI solutions should be human-centric, with people’s interests, well-being and safety as the primary consideration. It is deliberately algorithm-agnostic, technology-agnostic, sector-agnostic and scale-agnostic, so the same structure works for a bank, a hospital or a retailer.

Two companions sit alongside it. The Implementation and Self-Assessment Guide for Organisations (ISAGO) turns the framework into questions an organisation can answer about its own practices. A Compendium of Use Cases shows how organisations have applied it.

Who it applies to

Any organisation deploying AI at scale, in Singapore or elsewhere. It is voluntary, and the framework itself says that adopting it does not absolve an organisation from complying with the law. In practice it is the vocabulary Singapore regulators and many Asian customers use when they ask how an AI system is governed. MAS lists IMDA’s Model AI Governance Framework among the national initiatives financial institutions may refer to.

What it requires

The framework is organised into four areas.

Internal governance structures and measures. Clear roles and responsibilities for the ethical deployment of AI, adapted from existing governance where possible. AI risk can sit inside the enterprise risk management structure, ethical considerations can go through an ethics review board, and the people with AI governance roles should be properly trained. The framework calls the sponsorship and participation of top management and the board crucial.

Determining the level of human involvement in AI-augmented decision-making. This is the framework’s best-known contribution. An organisation weighs the probability of harm against the severity of harm to an individual or organisation and uses the result to choose one of three approaches.

• Human-in-the-loop. A person retains full control and the AI only recommends.
• Human-over-the-loop. A person monitors and can step in when the system behaves unexpectedly.
• Human-out-of-the-loop. The AI decides and acts with no human oversight of execution.

High severity and high probability point towards human-in-the-loop. Low severity and low probability can justify human-out-of-the-loop. The framework is clear that the definition of harm and how probability and severity are computed depend on context, and that the choice should be revisited as circumstances and societal norms change.

Operations management. The disciplines across the AI lifecycle. Data accountability covers understanding data lineage, ensuring data quality and minimising inherent bias in datasets. Algorithm and model practices cover explainability, repeatability, robustness, regular tuning, reproducibility, traceability and auditability.

Stakeholder interaction and communication. Telling people that AI is in use, having a policy on what to explain and to whom, considering an option to opt out, giving people channels for feedback and review, testing the user interface, and communicating in language people can follow.

Key dates

• 23 January 2019. First edition launched.
• 21 January 2020. Second edition and ISAGO released.

The framework has not been reissued since. IMDA has extended the approach through the Model AI Governance Framework for Generative AI (May 2024) and the Model AI Governance Framework for Agentic AI (January 2026, updated May 2026).

How it relates to other frameworks

The generative AI and agentic AI frameworks are extensions of this one. The agentic framework in particular adapts the human involvement idea to agents, adding significant checkpoints for human approval and guarding against automation bias. AI Verify turns the shared principles into process checks and technical tests.

Outside Singapore, the four areas map closely onto the management system clauses of ISO/IEC 42001 and the Govern, Map, Measure and Manage functions of the NIST AI RMF. The probability and severity approach to human oversight is a practical way to meet the human oversight expectations that appear in most frameworks, including the EU AI Act’s requirements for high-risk systems. Organisations that already run one of those frameworks will find most of the work done.

Working with it in Aicura

Singapore framework content is available in Aicura on the Enterprise tier. An administrator enables it in Settings under Frameworks, where the available frameworks follow the subscription tier. Risk and impact assessments then draw on it, including the human involvement calibration, and governance policy reviews can be set to apply it.

The AI Register already records each system’s and agent’s human oversight level, such as human-out-of-the-loop, which is the framework’s own vocabulary. Assessment output is guidance for your judgement. It does not certify a system or declare it aligned with the framework.

A note on this page

This is Aicura’s reading of the framework, written to help you understand it. It is not legal advice and it is not the framework itself. Read the primary sources from IMDA and the PDPC, and take advice on how Singapore’s expectations bear on your organisation.